Baryonification: An alternative to hydrodynamical simulations for cosmological studies

We present an improved baryonification (BFC) model that modifies dark-matter-only $N$-body simulations to generate particle-level outputs for gas, dark matter, and stars. Unlike previous implementations, our approach first splits each simulation particle into separate dark matter and baryonic components, which are then displaced individually using the BFC technique. By applying the hydrostatic and ideal gas equations, we assign pressure and temperature values to individual gas particles. The model is validated against hydrodynamical simulations from the FLAMINGO and TNG suites (which feature varied feedback prescriptions) showing good agreement at the level of density and pressure profiles across a wide range of halo masses. As a further step, we calibrate the BFC model parameters to gas and stellar mass ratio profiles from the hydrodynamical simulations. Based on these calibrations, we baryonify $N$-body simulations and compare the resulting total matter power spectrum suppressions to the ones from the same hydrodynamical simulation. Carrying out this test of the BFC method at each redshift individually, we obtain a 2 percent agreement up to $k=5,h$/Mpc across all tested feedback scenarios. We also define a reduced, 2+1 parameter BFC model that simultaneously accounts for feedback variations (2 parameters) and redshift evolution (1 parameter). The 2+1 parameter model agrees with the hydrodynamical simulations to better than 2.5 percent over the scales and redshifts relevant for cosmological surveys. Finally, we present a map-level comparison between a baryonified $N$-body simulation and a full hydrodynamical run from the TNG simulation suite. Visual inspection of dark matter, gas, and stellar density fields, along with the integrated pressure map, shows promising agreement. Further work is needed to quantify the accuracy at the level of observables.

💡 Research Summary

This paper introduces an upgraded baryonification (BFC) framework that transforms dark‑matter‑only N‑body simulations into particle‑level realizations of gas, stars, and dark matter. The key innovation is to split each simulation particle into a dark‑matter and a baryonic component before applying displacements, rather than moving the original particles directly. The displacement is computed radially around halo centers using a function that depends on halo mass, concentration, and a set of BFC parameters. The initial density field is modeled as a truncated NFW profile plus a two‑halo term; the final field is built from separate profiles for hot gas, inner cold gas, central galaxy stars, satellite stars, and the back‑reacted dark‑matter distribution.

Gas particles receive pressure and temperature values derived from hydrostatic equilibrium and the ideal‑gas law, enabling the generation of pressure maps alongside density fields. The model’s parameters are calibrated against high‑resolution hydrodynamical simulations from the FLAMINGO and Illustris‑TNG suites, which span a variety of AGN and stellar feedback prescriptions. Calibration is performed on radial gas‑to‑stellar mass‑ratio profiles rather than directly on power spectra, providing a more physically motivated constraint.

Validation proceeds on two levels. First, the BFC‑generated radial density and pressure profiles match those from the reference hydrodynamical runs to within a few percent across halo masses from groups to massive clusters. Second, after fitting the BFC parameters to the profile data, the total‑matter power‑spectrum suppression predicted by the baryonified N‑body boxes agrees with the corresponding hydrodynamical simulations to better than 2 % up to k ≈ 5 h Mpc⁻¹ for each redshift slice. This demonstrates that a profile‑based calibration can faithfully reproduce clustering statistics without direct power‑spectrum fitting.

To make the method practical for large cosmological analyses, the authors propose a reduced “2 + 1” parameter model: two parameters capture the amplitude of feedback‑induced gas and stellar redistribution, while a single redshift‑evolution parameter accounts for the modest change of these effects with cosmic time. This compact model retains sub‑2.5 % accuracy over the scales (k ≲ 5 h Mpc⁻¹) and redshifts relevant to upcoming weak‑lensing, galaxy‑clustering, X‑ray, and Sunyaev‑Zel’dovich surveys.

A visual comparison between a BFC‑processed box and the full TNG‑100 simulation shows that gas density fields are smoother but retain the large‑scale filamentary structure, dark‑matter fields are virtually unchanged, and stellar particles trace the high‑density peaks similarly in both cases. Minor discrepancies appear in the distribution of satellite galaxies and in small‑scale non‑spherical features, indicating areas for future refinement.

The paper also includes a treatment of the back‑reaction of baryons on the dark‑matter profile, showing that incorporating this effect improves the agreement of the power‑spectrum suppression. Overall, the BFC approach offers a computationally cheap (orders of magnitude faster than full hydrodynamics) yet accurate tool for generating multi‑probe mock catalogs, making it highly suitable for the analysis pipelines of next‑generation cosmological surveys. Future work will focus on extending the model to incorporate observational constraints directly, improving satellite‑galaxy modeling, and performing field‑level validation against real survey data.